FACTS NEVER MAKE DECISIONS.

The outcome is not the decision

A 55-year-old man has a heart condition. Surgery could relieve his pain and extend his life by five years. But 8% of patients die during the procedure. The physician decides to operate.

In 1988, Jonathan Baron and John C. Hershey ↗ presented two versions of this same case. The reasoning, expected benefits and stated risk remained identical; only the outcome changed. The decision received an average rating of +0.85 after success, compared with −0.05 after the patient’s death.

They had identified outcome bias: our tendency to judge a decision by what happened afterwards, even when that information could not have been known at the time of choice.

Annie Duke popularised a related distinction in Thinking in Bets ↗: under uncertainty, a good decision can produce a poor outcome, and a bad decision can succeed through luck. A decision never guarantees an outcome. It commits to a line of reasoning across several possible outcomes.

The first discipline is to ask before acting: “If the risk materialises, will I still consider this decision legitimate?” If the 8% mortality risk was properly estimated, understood and accepted, the patient falling within that 8% does not retroactively turn the decision into an error.

The “8%” must survive scrutiny

Stopping at risk acceptance would create the opposite bias: invoking bad luck to protect the decision-maker too easily. A poor outcome should be neither an automatic verdict nor an excuse. It should trigger an audit.

1. Does the death genuinely fall within the residual risk identified before the decision?

2. Was that risk estimated from reliable data, appropriate to the case and correctly interpreted?

3. Was a signal available at the time ignored, minimised or concealed?

4. Was the decision executed in accordance with the protocol on which the risk estimate depended?

If these checks confirm the original reasoning, the outcome is tragic, but the quality of the decision remains unchanged. If they reveal that the rate was wrong, relevant information had been excluded or execution departed from protocol, the judgement must change. Not because the patient died, but because the death exposed a defective representation of reality.

Who accepts the risk—and who bears it?

The casino analogy helps explain probability: before placing a bet, one must accept the possibility of losing. But the analogy conceals a question of power.

At a casino, the person making the decision generally risks their own stake. In an organisation, the decision-maker may impose the risk on others: employees, customers, shareholders or citizens. In the medical case, the physician cannot accept the 8% alone; the patient must understand the risk because the patient bears the irreversible consequence.

Evaluating a decision therefore requires four distinct judgements: the quality of the reasoning at the time, the actual severity of the consequence, the reliability of the risk model, and the responsibility of those who decided, consented, executed or bore the loss.

Confusing these four objects produces either a convenient culprit or organised irresponsibility.

Preventing the outcome from rewriting the past

Before any important decision, a robust organisation should preserve the available facts, selected assumptions, known uncertainties, alternatives considered, accepted risk and the conditions that would cause the judgement to be revised.

This record does not make the decision infallible. It prevents a future outcome from falsifying the actual state of knowledge at the moment action was required.

Facts choose neither the objective, nor the acceptable level of risk, nor the price to be paid. They never make decisions. They delimit the space of legitimate decisions.

The right question is therefore not only: “Did it succeed?” It is: “What did the decision rest on—and does what the outcome revealed justify changing that judgement?”

At that boundary, bad luck, error, negligence and manipulation can finally be distinguished.